【作者】 孙佳音； 李海峰； 王殿胜； 韩纪庆；

【Author】 Sun Jiayin Li Haifeng Wang Diansheng (Speech Processing Lab,School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001 Heilongjiang)

【机构】 哈尔滨工业大学计算机科学与技术学院语音处理研究室；

【摘要】 作为语音识别领域的核心热点技术之一,关键词检出技术(KWS)近年来得到了长足的发展。虽然目前在语音识别领域隐马尔可夫模型(HMM)占主导地位,但就关键词检出技术而言,人工神经网络(ANN),以出色的判别的能力,较小的计算量,更高的灵活度,成为重点研究的方向之一。本文提出将一种具有主副两个网络结构的新型神经网络--时间延迟集成网络(TANN)用于关键词检出,很好的解决了神经网络在处理时序分类问题上遭遇的困难,并且巧妙的回避了语音识别中的时间对正问题,并辅之以一种基于熵误差函数(EEF)的快速收敛且避免局部极小情况出现的网络训练算法。并通过初步实验证明这种网络在关键词检出上的性能表现与HMM相差无几,值得今后进行细致深入的研究。更多还原

【Abstract】 As one key research field and application hotpot of speech recognition,Keyword Spotting(KWS) technology has made a significant improvement in recent years.Although Hidden Markov Model(HMM) has been mainstream of speech recognition for years,Artificial Neural Network(ANN),with its strong discriminatory ability,low computation cost,high flexibility,has become an efficient solution to KWS.In this paper,we propose an investigation into a novel strategy on KWS,in which a new type of time-delayed neural network with a pair of networks called Time-Accumulation Neural Network(TANN) are adopted.TANN is quite a solution to the problem ANN faced in temporal sequence pattern classification,and time warping in speech recognition. Furthermore,a new network training algorithm with entropy error function(EEF) is used to train the TANN,this algorithm is very efficient with no local minimum.Preliminary experiments show that the performance of the proposed strategy based KWS can reach the performance of HMM based ones at a relatively computational consummation,and proved our strategy to be quite promising.更多还原